Reference · Terminology
Glossary of Poker AI and Detection Terminology
This entry defines the vocabulary used throughout the present study and the wider poker-AI literature: solver theory, opponent modelling, network architecture of iPoker, and the detection apparatus observed on the Bet365 skin. Definitions are written in a citation-style register and are intended to be referenced from the primary document.
- GTOGame Theory Optimal
- A strategy that cannot be exploited by any opponent because it plays a Nash-equilibrium frequency distribution. A solver-anchored bot uses GTO output as its baseline before layering exploitative deviation.
- CFRCounterfactual Regret Minimisation
- The iterative self-play algorithm used to approximate Nash equilibria in imperfect-information games like poker. Pluribus and Libratus are CFR-derived. Convergence produces the action-frequency tables a bot queries.
- Solver
- Offline software (PioSolver, GTO+, MonkerSolver) that computes a near-equilibrium strategy for a defined game tree. The output is a large lookup table keyed on game state; the solver is the theory book, the bot is the player using it.
- HUDHeads-Up Display
- An on-screen overlay showing opponent statistics gathered from hand history. Third-party HUDs are prohibited across iPoker skins and blocked by client-side process detection, which pushes modern bots toward online opponent modelling instead.
- Rake
- The commission an operator takes from each pot or tournament entry. iPoker rake runs 5–7% on low-stakes NLH cash dropping to 3–4% at mid-stakes, and is the main constraint on which stakes are bot-viable after costs.
- RNGRandom Number Generator
- The system producing card order. Modern poker uses a cryptographically secure pseudo-random number generator (CSPRNG) seeded from multiple entropy sources, committed server-side before any card information reaches a client.
- CSPRNGCryptographically Secure PRNG
- A random number generator whose output cannot be predicted even given prior outputs. The reason RNG-prediction "hacks" do not work: poker exposes outputs far too slowly to recover the generator state.
- eCOGRA
- An independent testing and certification body that historically attested the iPoker RNG. Current attestations are jurisdiction-specific (UKGC, MGA, and others). An audit confirms the audit happened, not that the system is unbreakable.
- Multi-tabling
- Playing several tables at once. Trivial for software and a primary economic reason bots exist; also a detection vector, since sequential decision-clocking across tables can leak cross-table action-latency correlations.
- ICMIndependent Chip Model
- A model converting tournament chip stacks into real-money equity, critical for bubble and final-table decisions. Solver coverage of ICM spots is patchy, which is why MTTs remain a hard automation target.
- Fingerprinting
- Building a stable identifier from device and behavioural signals — input timing, mouse-path geometry, touch dwell, sensor data, idle behaviour. Playtech joins behavioural fingerprints across every iPoker skin on its internal player ID.
- VPIPVoluntarily Put In Pot
- The percentage of hands a player voluntarily enters. A core opponent-model statistic; pure-GTO bots cluster at population mass with abnormally low variance, which is itself a detection signal.
- PFRPre-Flop Raise
- The percentage of hands a player raises before the flop. Paired with VPIP to classify playing style and feed both opponent modelling and statistical play-pattern detection.
- 3-bet
- The third bet in a betting sequence (a re-raise). 3-bet frequency by position is a standard feature in both solver strategy and the detection classifier.
- C-betContinuation Bet
- A bet on the flop by the player who raised pre-flop. Fold-to-cbet by board texture is a high-signal statistic for both exploitation and detection.
- Spin / Twister
- iPoker’s lottery-style three-handed hyper-turbo (10bb start) with a random prize multiplier up to 1000x the buy-in. Push-fold trees are well-solved, but multiplier variance dominates the EV signal over tens of thousands of games.
- Speed Poker
- iPoker’s fast-fold variant, comparable to Rush & Cash or Zoom. Players are reseated immediately after folding, which breaks long-horizon opponent modelling faster than a static table.
- RTAReal-Time Assistance
- Using solver or AI output during a live hand without full automation. Prohibited; detected through statistical play-pattern analysis over volume rather than client telemetry.
- Adversarial classification
- The formal frame for anti-detection: the operator builds a classifier separating bot from human behaviour, and the bot produces a distribution the classifier cannot separate from the population while preserving EV. Dates to Dalvi et al. (2004) and Lowd & Meek (2005).
- KYCKnow Your Customer
- Identity- and source-of-funds verification a regulated operator must perform. Bet365’s sportsbook-grade KYC is heavier than at a pure poker operator and closes the multi-accounting graph cleanly.
- Skin
- A branded operator front-end on a shared liquidity network. Bet365 Poker is a skin on Playtech’s iPoker Network; player pool, RNG, and core anti-cheat are shared across all skins.
- Chip dumping
- Deliberately losing chips to transfer value between colluding accounts. A top operator priority under UKGC and MGA anti-collusion requirements, caught via the account graph plus suspicious hand sequences.
- State abstraction
- Collapsing the game state into equivalence classes (board texture, stack-to-pot ratio, position, action history) to shrink a solver lookup table. Combined with action abstraction it compresses CFR output by four to five orders of magnitude with bounded EV loss.
- Action abstraction
- Discretising bet sizes to a finite set (33%, 50%, 66%, 100%, 150%, all-in) so the solver solves only at those points, making mobile-budget inference feasible.